The Logic of Recursive Equations Author(s): A. J. C. Hurkens, Monica McArthur, Yiannis N. Moschovakis, Lawrence S. Moss, Glen T. Whitney Source: The Journal of Symbolic Logic, Vol. 63, No. 2 (Jun., 1998), pp. 451-478 Published by: Association for Symbolic Logic Stable URL: http://www.jstor.org/stable/2586843 . Accessed: 19/09/2011 22:53 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Symbolic Logic. http://www.jstor.org THE JOURNAL OF SYMBOLIC LOGIC Volume 63, Number 2, June 1998 THE LOGIC OF RECURSIVE EQUATIONS A. J. C. HURKENS, MONICA McARTHUR, YIANNIS N. MOSCHOVAKIS, LAWRENCE S. MOSS, AND GLEN T. WHITNEY Abstract. We study logical systems for reasoning about equations involving recursive definitions. In particular, we are interested in "propositional" fragments of the functional language of recursion FLR [18, 17], i.e., without the value passing or abstraction allowed in FLR. The 'pure," propositional fragment FLRo turns out to coincide with the iteration theories of [1]. Our main focus here concerns the sharp contrast between the simple class of valid identities and the very complex consequence relation over several natural classes of models. In [18, 17], Moschovakis introduces the language FLR to study general recursive definitions of the form p(u) f (u,p). The functional f determines how to compute values of the function (or "program") p based perhaps on other values of p. A key special case consists of simple fixed- point equations P f (P), in which the dependence of p on some "input" has been eliminated or suppressed. Therefore, we investigate here two "propositional" fragments of FLR: FLRO (first introduced as the language Y in [19]) and ELR. Completeness questions for FLRO and ELR are already far from trivial, and the simpler language makes broader classes of models more accessible. To briefly summarize, FLRO takes a primitive stock of variables and function symbols, and it forms functional terms in the usual way and recursionterms by the construction (1) B where {xl = Al, x11= A1l1. The variables xi typically occur in the terms Ai, so in fact FLROcan express systems of simultaneous fixed-point equations. Definitions of this kind occur in logic and computer science in a wide variety of settings. Of course, each time a term of the form (1) is used, one must be able to find well-defined semantics. (This is to avoid terms like x where {x = 1 + x} or R where {x E R +-* x V R}.) The most common general way to provide semantics Received October 4, 1996. During the preparation of this paper, Whitney was supported by the Fannie and John Hertz Foun- dation and by a grant from the National Science Foundation. ? 1998, Association for Symbolic Logic 0022-481 2/98/6302-0009/S3.80 451 452 HURKENS, McARTHUR, MOSCHOVAKIS, MOSS, AND WHITNEY for fixed point terms is by appeal to some sort of result on fixed points of monotone maps on a directed-complete partial order (cpo). This covers many of the cases for computer science. On the other hand, fixed points arise in many other contexts, and for this foundational study we present a very general notion of structure for FLRO. In particular, we wish to admit "intensional" models, in which distinct functions may take identical values at all points of a structure but be assigned different fixed points; this type of intensionality is at the core of many studies of concurrent computation. Early attempts to understand and prove properties of recursivedefinitions in var- ious specific contexts include [5, 6, 15, 13]. The general study of recursion equations has been pursued under several guises since then:I as recursiveapplicative program schemes [4], u-calculus [2], and perhaps most notably as the iteration theories [10, 1] of Bloom and Esik. The latter work builds on Lawvere'sintroduction of algebraic theories [14] in order to get categorical presentations of universal algebra, and El- got's use of these in connection with flowchart schemes [7]. Thus, the relationship of this paper and [1] is roughly that between ordinary equational logic and algebraic theories. In particular, we will provide an explicit "dictionary" showing that the categories of FLR0-structures and iteration theories are equivalent, so that answers to many basic questions can be read off from known results in the iteration theory context. On the other hand, we present new results and questions suggested by the logical formulation, and this presentation will hopefully make the subject more accessible to those with a mathematical logic background. ?1. Elementary formal language of recursion. 1.1. Syntax. Fix a countably infinite set {Vl, V2, V3,.. .} of variables. A signature r is a ranked set of functionsymbols; in other words, each symbol f has an associated arity, the (nonnegative integer) number of formal arguments it will take. Write l,? for the subset of r consisting of n-ary symbols. The following induction defines the terms of the language FLRo (). (1) Any variable x is a term by itself. (2) f (E1, ... I,E,) is a term if f E la and El through En are terms. (3) Eo where {xl = El,... , x, = El} is a term for any distinct variables x1, .. ., xn and any terms E0, . ., El. Intuitively, the second clause corresponds to function composition, and the third clause gives syntax for the solution of systems of recursiveequations. The following expression schematically summarizes the whole definition: E := x I f (El, . .., En1)I Eo where { xi = El, I .. I x17= En1 Syntactic notions concerning FLRo are defined as usual, including closed and open terms, substitutions, free substitutions, and fresh variables. The where in clause 3 binds variables xl through x,. We fix some notational conventions. The plain symbol g will abbreviate go for nullary function symbols (i.e., constant symbols). The special symbol I will I Another approach to axiomatic theories of recursion, algebraic recursion theory [21, 11, 12], has a different emphasis: it confines itself to least (or initial) fixed points, and seeks to understand the additional combinatorial properties needed to support stronger results analogous to classical recursion theory. THE LOGIC OF RECURSIVE EQUATIONS 453 abbreviatethe term x where {x = x } (the exact choice of variable is irrelevant). In- formal vector notation will be used throughout, e.g., E where {x = Al abbreviates E where {xl = A1, ... , x, = A, } . When s is a function from variables to terms, we write E[s] for the result of substituting the term s (x) for the free occurrences of x in E, for each x in the domain of s. Sometimes the substitution s may be displayed explicitly, e.g., E[M/x*l] denotes the result of substituting the FLROterm Mi for xi, for each xi in the sequence x*. Further, if the term E has been written as E(xl,. , Xn), displaying (some of) its free variables, then the substitution E[M/lZ] may also be written E (M1, . ., M,,) . Finally, A _ B means that the expressions A and B are identical. Alphabetic variants. Suppose - is a relation on some variables. Define A B to mean that A and B are syntactically identical up to '-, i.e., that there is some C(ZI, . , Zn) and lists of variables x1,... , x, and yi, . , y, (which may have repetitions) such that A _ C (x,I ... , xl), B C (y1,. , y),,), and xi ,- yi for all i. Formulas. If A and B are terms, then for any sequence of distinct variables x (in- cluding the empty sequence), Vx (A = B) is a formula of FLRO(r). Intuitively, the formulas will be used to express equations, such as f (x) = g (y), which might hold for some particular x and y, and identities, like Vx (f (x) -= (f (x))), expressing the idempotence of f. This Vx`isanother variable-bindingoperator: all occurrences of variables from x-which are free in A and B are bound in Vx (A = B), and all other occurrences remain free or bound as they were. A closed formula will be known as an identity. Substitutions apply to formulas as a whole; Vx (A = B)[M/J] replaces free occurrences of the variables in the list V wherever they occur in A or B, but of course this substitution will not replace any occurrences of variables from x. If bVx` (A = B) is a formula, then Vy (qi) is an abbreviation for Vy, X (A = B), so that it makes sense to quantify any formula universally. 1.2. Poset semantics for FLRo. Interpreting the variables of FLRo as ranging over the elements of a poset D leads to three natural classes of semantic structures which will be central throughout this paper. The full definitions follow, but briefly the classes are Cont, in which D is complete and the functions are continuous; Mon, in which D is complete and the functions are merely monotone; and Wk, the "weak" structures in which the poset may not be complete but nevertheless still contains "enough" fixed points to interpret FLRo.2 First, let D be (directed-)complete. Given a signature r, choose a monotone function fA: D'2 -> D for each f in rc.